Here was the prompt to Gemini to create the text>audio script, developed after trial and error.
Disclaimer: the prompt will never produce the exact same result, and is influenced by other “learnings” the agent has done with related work.
The final output script MUST begin *exactly* with the following paragraph, without any preceding text or modifications:
“{FIXED
_
INTRO
_
TEXT}”
Following that exact paragraph, continue with the rest of the newscast as specified below.
Objective: Create a plain-text news script suitable for text-to-speech (TTS) delivery, adhering to specific stylistic and content requirements.
I. Core Script Structure & Content:
1. Date of Broadcast (for the newscast’s own intro, which follows the fixed intro above):
{formatted
_
broadcast
_
date
_
full}
2. Source Material (all stories are from {formatted
_
broadcast
_
date
_
full}):
You will be provided with a block of raw news reports. Select, synthesize, and rewrite stories from this material. Do not simply copy sections; rephrase for a newscast style.
“`
{source
_
material
_
text}
“`
3. Number of Stories & Length:
Target 5 to 6 distinct news stories selected from the provided source material.
Each story should be approximately 45 seconds in spoken length.
The total script (including the fixed intro about Leslie Katz provided at the very beginning of this prompt) should be approximately 5 to 5.5 minutes in total spoken length.
II. Script Format & Style Requirements:
{“”} # Removed “1. Title (of the newscast segment): Bay City News” to prevent it from being spoken.
2. Newscast Introduction (this comes *after* the fixed intro text provided at the very beginning):
Template: “Hello, and welcome to Bay City News for {formatted
_
broadcast
_
date
_
intro}.
Here is a look at some of the top stories from across the region.”
3. Story Presentation:
Conciseness: Use clear, concise sentences. Avoid jargon or overly complex sentence structures.
Factual & Neutral: Maintain a neutral, objective, and factual tone. Avoid opinions or sensationalism.
Attribution: Attribute information where appropriate (e.g.,”police said,” “according to a statement from X,” “officials announced”).
4. Transitions:
Implement smooth transitions between each news story.
Examples:
“From that news, we turn to…”,
“Meanwhile, in [Location]…”,
“Shifting our focus to…”,
“In other developments…”,
“Finally, some news on…”
5. Phonetic Cues for TTS (Apply VERY CONSERVATIVELY):
Purpose: To guide TTS pronunciation only for names or places genuinely and significantly likely to be mispronounced by a standard, high-quality TTS engine.
Selection Criteria: Use your best judgment very conservatively. Default to standard English spelling for all names and places unless there is a high probability of mispronunciation by typical TTS. Prioritize only truly non-obvious non-English names or English names with exceptionally unusual or ambiguous pronunciations that standard TTS would almost certainly get wrong (e.g., Siobhan, Niamh, a very obscure or foreign place name not commonly known).
Crucially, if a common English name (e.g., Michael, Jennifer, David, Gavin, Leslie, Jason, Daniel, Ray, Scott, Jerry) or a well-known English place name (e.g., London, New York, San Francisco, Sacramento, San Mateo, Rohnert Park) is likely to be pronounced correctly or acceptably by standard TTS, YOU MUST USE ITS STANDARD SPELLING. Do not use phonetic spellings for these common cases. The goal is to assist the TTS only when absolutely necessary, not to phoneticize most proper nouns. If in doubt, err on the side of standard spelling.
Format:
The phonetic spelling should replace the original spelling entirely.
Use lowercase letters only for the phonetic spelling.
Use hyphens to separate syllables (e.g., wah-keen, ty-mee-shuh see-mor). Note:
Capitalize syllables only if they are meant to be stressed significantly more than usual for a proper name, otherwise default to all lowercase.
Do NOT include the original spelling alongside the phonetic cue.
Do NOT use parentheses around the phonetic cue.
Example of correct usage (very sparingly):
“…awaits a decision from ty-mee-shuh see-mor…”
“…near the San wah-keen River…”
Example of incorrect usage (over-phoneticization to be avoided):
“…gav-in noo-suhm announced…”
(Should be ‘Gavin Newsom announced…’ as standard TTS will likely handle this well).
“…lez-lee kats said…”
(Should be ‘Leslie Katz said…’).
6. Abbreviations:
Avoid unexplained abbreviations.
Define any abbreviation on its first use (e.g.,
“Peninsula Open Space Trust, or POST,” then subsequently use “POST”).
7. Dates within Stories:
Spell out all dates fully within the story content (e.g., “May 10, 2025,” “Monday, May seventh”).
8. Character Set:
Use only standard ASCII characters.
9. Newscast Outro:
Conclude with a short, professional outro.
Template: “And those are some of the top stories we’re following. Thank you for joining us for Bay City News.”
10. Delivery Format:
The complete script must be delivered as a single block of plain text, ready for input into a text-to-speech engine.III.
Process Notes for AI:
Prioritize stories that are most newsworthy and relevant to the specified broadcast date from the provided source material.
If source material provides multiple updates on the same event, synthesize them into a single, coherent story reflecting the latest available information for
{formatted
_
broadcast
_
date
_
full}.
Ensure a good narrative flow throughout the entire newscast.
Adhere strictly to the conservative phoneticization guidelines in II.5. Standard spelling is the default for almost all names and places.